activemq-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From franz1981 <...@git.apache.org>
Subject [GitHub] activemq-artemis issue #1119: ARTEMIS-1025 OutOfDirectMemoryError raised fro...
Date Sat, 01 Apr 2017 06:17:56 GMT
Github user franz1981 commented on the issue:

    https://github.com/apache/activemq-artemis/pull/1119
  
    @clebertsuconic Exactly, it tries to block it for a finite amount of time and only if
not in the event loop: you've mentioned about stalling behaviours and I've dig in the Netty
code to find out a possible reason of it, finding only 1 answer: maybe we share the I/O threads
(via a pool?) with Netty and calling ``writeAndFlush`` while on one of them.
    Altought it is considered legal, IMHO is something that must be avoided because it increases
the risk to go OOM (it steals time to a thread that would be happier while doing I/O stuff
and releasing the sent ``ByteBuf``s) and it will lead to unexpected infinite stalls...
    It could be the key to solve other strange blocking behaviours happening on replication
too!
    
    Anwers:
    i - The original behaviour was already to perform a ``writeAndFlush`` there, but without
applying any backpressure on the caller, streaming huge messages with clusters (with 1 GB
per msg) would lead anyway to log some warns but then going OOM.
    As you said we need to perform flow control in our codebase, but the granularity allowed
using Netty in idiomatic way couldn't be enough: a "writable" flag is good for the most cases
but not enough with very high traffic with long burst of data (=== streaming huge messages
in batches).
    Maybe we need an API change, like returning a ``boolean`` on the ``NettyConnection::write``
method, in order to let the caller choose what to do as part of its duties, but is weird (
as a user, I mean).
    Monday we could talk about it better or I could ask on the forum for feedbacks: solving
it properly will improve Artemis a lot, I'm sure of it :+1:  
    ii - :100: No more crazy properties
    
    Thanks Cleb!!!



---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---

Mime
View raw message